F-1332 L字接続線分による物体の位置・姿勢推定(J29-1 ロボットビジョン)(J29 センサ・アクチュエータシステムとその知能化 : 実環境で活躍するメカトロニクスをめざして)

概要

論文の詳細を見る
Authors examine an object's pose estimation method from monocular input images: by matching L-shaped line segments between 2-D input images and 3-D objects, the pose, i.e., orientations and positions, is obtained. The L's are intentionally formed, adjoining two line segments both on the images and on the object. However, since the method provides up to four solutions, we should select the best solution among them. This paper describes a method for the selection. Calculating the 3-D coordinates of the object's gravity centers for all the solutions of each L, we can select a solution that achieves the best agreement through L's. Some practical experiments were conducted to show an effectiveness of the proposed selection method.
一般社団法人日本機械学会の論文
2001-08-22